Artificial intelligence-assisted disease diagnosis: A narrative review of diagnostic advances and challenges toward clinical integration.
Authors
Affiliations (3)
Affiliations (3)
- Hebei Key Laboratory of Analysis and Control of Zoonotic Pathogenic Microorganism, College of Life Sciences, Hebei Agricultural University, Baoding, China.
- Clinical Medical Centers, China National Biotech Group Co., Ltd, Beijing, China.
- Institute of Hemu Biotechnology, Beijing Hemu Biotechnology Co., Ltd, Beijing, China.
Abstract
Artificial Intelligence (AI) is increasingly being applied in disease diagnosis and has shown considerable potential for improving diagnostic accuracy, efficiency, and accessibility. This narrative review aims to examine recent advances in AI-assisted disease diagnosis from the perspective of the clinical diagnostic pathway, with particular attention to how AI is applied across successive stages of disease diagnosis and clinical practice. It synthesizes recent advances in AI-assisted disease diagnosis from representative studies. This review is structured around the diagnostic pathway, encompassing disease detection and classification, risk stratification and prediction, and finally, clinical integration. The article critically examines the role of AI in three core areas: (1) the application of multimodal data for the detection and classification of non-communicable diseases, including neurological disorders and cancer; (2) genetic risk stratification and diagnosis of Mendelian disorders; (3) advancements in medical image analysis through multimodal fusion, radiomics, and real-time assistance. A central theme of this review is that AI is evolving from a single-modal detection tool toward broader clinical applications, with current evidence supporting its role primarily as a decision-support technology rather than as a replacement for clinical judgment. However, this transition is constrained by significant challenges, including data quality, model generalizability, clinical validation, and ethical and regulatory hurdles. By synthesizing these advances and limitations, this review aims to provide clinicians and researchers with a balanced perspective, thereby offering insights into the development of a trustworthy, next-generation diagnostic ecosystem.